Plasma cleaning for improved wire bonding on thinfilm hybrids Plasma cleaning removes surface contamination. When assembled hybrids are cleaned just prior to wire bonding, energy input and gas flow rate must be closely' controlled. By H.B. Bonham and P.V. Plunkett, Rockwell International, Richardson, Texas Many authors have discussed the influence of surface contamination upon the bondability of hybrid metallizations.' These discussions show that contamination reduces production throughput since repeated attempts may be required to achieve an intraconnection, and show that contamination reduces bond reliability since bonding is only achieved between islands of contamination. Traditionally, hybrid manufacturers use incoming tests, both visual and bondability, to certify component bondability prior to circuit assembly. The problem with this approach is what to do with parts that fail incoming, particularly when manufacturing schedules are short. Cleaning is obviously required. Wet chemical techniques, which are usually employed, 'are &%gned to ligfitly etch metallic bond pads or to dissolve inorganic contaminants. These techniques are expensive since hybrid components are small and difficult to handle, and are frequently ineffective. As a last resort, many hybrid manufacturers use nondestructive pull tests to screen intraconnection in an attempt to identify and remove components which are contaminated. Nondestructive pull testing is only margnally successful since nondestructive force limits are low. Also for gold to aiuminum thermocompression ball bonding, the asmade bond strength may be high but strength may be low after limited temperature storage. (The temperature storage referenced here is less than that required for Kirkendall failure.) Dry chemical plasma cleaning is a method that offers a solution to these problems. The method does not require special fixturing, since completed hybrids may be cleaned just prior to wire bonding. The idea is to use RF power to generate free oxygen radicals. The radicals react with hydrocarbons to produce water vapor, carbon monoxide, and carbon dioxide, which are pumped from the system. Thus, bonding surfaces are cleaned of organic contamination by chemical oxidation. Since hybrids employ a number of oxidizable materials (epoxies, silver, and tantalum nitride resistors), the concentration of oxygen radicals and plasma clean time must be carefully controlled. Definitions and initial experimentation The objective of this study was to determine a parametric envelope for plasma cleaning of aluminum surfaces prior to thermocompression wire bonding. A parametric envelope illustrates the range of process parameters that yield acceptable cleaning from a bondability point of view without degrading adjacent hvbrid materials. le1 chambers. 4 in. in diameter and 8.5 in. long. Each chamber can easily accommodate 15 hybrids. Automatic control is built into the unit so that once the fundamental parameters are preset, the operator only loads and unloads the chambers. Since the procedure developed requires two cleaning steps, the plasma stripper was modified to handle two gases. oxygen and argon. Gases were switched using a manual valve. Prior to performing the experiment. it was necessary to define bondability and to select samples. Bonding was.. accomplished using conventional thermocompression ball bonding techniques with the following parameters: 1) 0.007in. diameter gold wire 2) Dwell time1.0 s 3) Base temperature220 C 4) Capillary temperature450"c 5) First bond force40 g 6) Second bond force30 g Bondability was evaluated by monitoring bond yield and nondestructive pull test yield and by measuring intraconnection pull strength after an accelerated life test. Bond yield is the percentage of bonds achieved on the first attempt. Nondestructive pull testing was done with the loop technique to a 1.0g level. The accelerated life test consisted of storing bonds at 300 C for 4 h. After the temperature stress. bonds were required to be k Modei PDS302 dry plasma strip stronger than the intraconnection per manufactured by LFE Corpora*% wire: that is. only wire failures were tion was used.' The unjt has two paral accepted. When the 1.0eV activation
. e:iergy shown hy Peck' wan used. the ;Itw*C. strew i. rquivulent to approximatrly 20 years nt W"C ambient. Sumlile selection was difficult since. the contamination levels of interest art! typically not visible with standard light optics. In addition, the level of contamination throughout nn incoming group of components typically varies. For most samples evaluated. no more than one bond in 100 would fail the bondability criteria. Although this failure rate is unacceptable for manufacturing. it is much too low for evaluating cleaning techniques, since extremelv large samples would be required. A large number of devices was found that bonded poorly. These devices were stored in air for a period of over three years and were initially stored because they demonstrated poor bondability. Visual inspection'of the parts did not show obvious contamination, although 30% of the bonds failed bondability testing. After bond samples were selected and bondability requirements were defined. samples were cleaned and bondability was evaluated. This initial study indicated that bondability of aluminum surfaces could be remarkably improved. but that silver metallization (epoxy filler and capacitor terminals) was discolored. The discoloration was only visual since capacitor terminals remained bondable. Samples of the discolored capacitors were baked at 300 C. This temperature should readily reduce silver oxide. It was found that the oxide was reduced within a few minutes. Lower temperatures were also tried. but the time needed to restore the metallic color was excessive: approximately 1 h at bonding temperatures of 220 C. Finally, the plasma system using argon in addition to oxygen was tried. The idea was to use the inherent heat generated in the cleaning to disassociate the silver oxide. Even though temperatures generated by this technique are lower than the temperatures evaluated above, the cleaner applies heating in a vacuum and bombards the oxide with argon. The study showed that the argon plasma easily restores the metallization color....... Generation of parametric envelope As stated earlier, the parametric envelope for plasma cleaning is an illustration of the range of cleaning parameters over which adequate cleaning, from bondability point of view, may be achieve'd. The envelope shows directly the interdependency of process parameters. Its boundaries indicate failure modes. Plasma cleaning as' defined above has six variables, three for the oxygen cleaning cycle and three for the argon cycle. The variables are: (I) process time, (2) gas flow rate, and (3) RF power. To simplify equipment operation, gas flow rates and the RF powers for oxygen were also used for argon. In addition, argon plasma time was kept constant at 2 min throughout the evaluation. First. bondability was evaluated in terms of oxygen process time. Approximately 60 bonds were produced for each process time evaluated. Bonds were evaluated by recording bond yield in percent, percent of ball lifts during nondestructive pull. and percent of ball lifts after the accelerated life testing. As seen in Figs. 1 and 2, plasma cleaning greatly improved bondability. For oxygen cleaning times of 9 min or greater, ceramic capacitor silver terminations were visually degraded. The first step in generating the 2. Infience of oxygen /low rate. T t "t A 20 i / OXYGEN PLASMA TIME! /MINUTES1 envelope was to determine the cleanlinesscompletion relationship; that is, to determine the combination of RF power and cleaning time that will result in aluminum surfaces sufficiently clean for thermocompression bonding. Gas flow rate was kept constant at 130 scc/min for this evaluation. The procedure consists of selecting a cleaning time and increasing RF power until completion is obtained. Completion here is defined as 100% wire failures after the 300 C temperature stress. 0% nondestructive pull test failures, and 100% bond yield. After determining the completion relationship for 13Oscc/ min gas flow, the relationship was verified for gas flow of 150 and 170 scc/min. Figure 3 illustrates the results. The RF powers above 150 W and below 50 W were not evaluated. Above 9F POWER '0 WATTS CLEANING 1lUE 5 MINUTES I 4CCELERATED LIFE & BOND YIELD 0 PERCENT NONDESTRUCTIO'I FAILURES 3 50 2 0?50 3w OXYGEN FLOW RATE SCC MlNUlfC t.
... k.. 150 W. nietnllic tilnix could be $puttcyed: beiow 50 W. clenning times would reduce production throughput. Since piasmu cleaning process time is dependent upon the number of parts or the level of part contamination. the data in Fig. 3 wns garhered in a full load condition. Full loading was simulated by placing a 2 by?.in. subgtrate coiited with photoresist in each chamber during cleaning. This quantity of organic contamination shouid be greater than the level of hybrid contamination in a cleaning run. By employing the techniques discussed. minimum cleaning parameters required to achieve bondability were established. To generate a parametric envelope. maximum cleaning parameten are also required. Samples were cleaned with a process time of 10 min and RF powers of 50, 100 and 150 W. All samples evaluated demonstrated good bondability: however, silver surfaces were visually degraded. The envelope shown in Fig. 3 illustrates the combinations pf plasma cieaning parameters effective for cleaning aluminum and gold metallizations if silver is not present. A wide range of cleaning parameters may be used. The envelope applies to both gold and aluminum, since the bonding investigated consisted of a ball bond to aluminum and a wedge bond to gold. Since silver metallizations are used in most hybrids. the parametric envelope of Fig. 3 typically may not be employed. This restriction is a result of choosing to implement plasma cleaning at prebond for cleaning assembled hybrids. A silver degradation relationship is required. To obtain this relationship, samples were cleaned at constant process time while the RF power was increased until visual damage appeared. Visual damage was determined using 1oOOX scanning electron micrographs. Any appearance of a cauliflower structure was cause for rejection. All silver samples cleaned met the defined bondability criteria. The silver degradation relationship is shown along with the completion relationship in Fig. 4. The data for this relationship was gathered under mini. mum loading conditions to?epre&nt' the worst case. Figure 4 illustrates the parametric envelope for plasma cleaning of assembled hybrids. Insufficient cleaning and overcleaning regions are illustrated, as 1s the regon of usable parameters. Two additional curves are shown in the figure. These curves represent constant energy. The first (12,OOO.I) closely follows the completion curve and the second (24,000 J) corresponds to the over clean OG silver degradation curve. The data. there 3. Pommetnc rntvlope /or aluminum and gold mcta Uizationa. 4. Pammetric envelope for piasma hybd cleaning.. fore, indicate that plasma cleaning may be evaluated only in terms of gas flow and the energy applied to the system. Selection of operating parameters After defining the parametric envelope, a set of opefating parameters was selected to minimize the effects of parameter variations. This operating point would be the center of the envelope if both RF power and cleaning time could be controlled with equal.,. I I.?. A A I.. A 6 mi EVALUATED 0 IO OXVGEN PLASMA IlUE imlmtesl c l v 220 SCC.'MINUTE GAS FLOW IYO SCCiMINUTE FLOW A I30 SCC:MINUTE GAS FLOW REGION OF VISUAL DAMAGE ' TO SILVER METALIUTIONS I I I I 1 I I I 2 3 4 5 ' 6 7 8 OXYGEN FIASMA TIME WNUTESI tolerances. Since cleaning time is more readily controlled. the RF power (the point in the upper left, 3.5 min. 85 W, of Fig. 4) was selected. Cleaning or process time may be controlled to z 10 scc/min. Thus, the width, height, and depth of the envelope through the operating point are: 1) 20 cleaning time toleranceswide 2) 10 RF power toleranceshigh. 3) 9 gas flow tolerancesdeep. Such an operating point should be easily controlled in manufacturing. I
Cwtitication of ptrmmetric envelope C't*rti!ication 01' the envelope was accomplished hy iis.wnii,ling and cleaning test simples at each envelope vertice..ilso. ;I control group with no cletining was tested. After cleaning, the simples were bonded and stressed ai :I(O"C' for 4 h. The stress was applied onlv to goldaluminum bonds, since temperature exposure of gold SI '1 ver bonds improves bond strength. Each test sample WPS pulled to destruction using the loop technique. Failure mode. and strength were recorded. Figure 5 illustrates the results for aluminum. This figure is a probability plot in which normal distributions are straight lines.' The envelope about these curves shows 99% confidence limits. This data clearly shows that goldaluminum bonding was greatly improved by plasma cleaning. The goldsilver bond sample predicts no more than one bond in 10,OOO having strength less than 1.3 g. For the goldsilver case. all interconnections tested failed as wue breaks. Plasma cleaning of tantalum nitride resistors Tantalum nitride resistors are stabilized using a hightemperature bake in air. The temperature soak anneals and forms an oxide that reduces further oxidation during circuit application. When plasma cleaning is used, it is possible to increase the oxide layer on resistors and thus force an increase in resistor value. Also. plasma cleaning may influence longterm resistor stability. Resistors were measured before and after plasma cleaning and after longterm stability stressing. A 0.026% increase in resistance resulted from plasma cleaning. Resistor stability was investigated using the standard 1,OOO h. 150 C stress in air. Percent change in resistance was determined after 160, 760, 1.O00, and 2.000 h. For each test, eight resistors were measured. Resistor stability was determined for resistors having 0, 1 h, and 2 h of stabilization bake. A plasmacleaned lot and a control lot (no cleaning) were evaluated. Plasma dt%mingc~mseted of 100 W of RF power, 15 min of oxygen plasma time, 2 min of argon plasma time, and 350scdmin gas flow rate. Thus. a worstcase condition was evaluated and the results are shown in Fig. 6. Resistor stabilitv was improved for the 0 and 1 h stabilization bakes: but for the 2 h stabilization bake, resistor stability decreased. Since resistor stabiiity is specified at 0.58 after 1,oOO h at 150OC in air. plasma cleaning did not effectively degrade the resistors. Identification of contamination Auger electron spectroscopy (AES) was utilized to determine the type and level of contamination on cleaned and uncleaned samples. Table 1 summarizes the relative AES peak intensities for the two cases. The oxygen peak was used as the reference peak because of its sensitivity to AES and its abun. 6. Resistor drift after plasma cleaning. I 7 */ c dance on the samples analyzed. The table also shows the relative peak intensities ~ after removing approximately 15 A of the cleaned sample and 60.A of the uncleaned samples. In addition, Auger analysis was also performed off the bond pad area to determine the extent of contamination. c CONlDOCNOCLEANING 4ASUA CLEANED
I Table 1. Relative AES Peak Intensities. I Uncleaned bond pads loo 330 6.4 5 5 3 a trace Plasma cleaned bond pods 100 250 11 3 5 4 4 Ion etched uncleaned Sam 100 326 11 112 Pie Ion etched cleaned sample 100 298 4 a6 Off bonding pads tvpical of 100 3 3 105 cleaned and uncleaned samples The uncleaned samples revealed a large amount of carbon with a trace of nitrogen detected. This is indicative of photoresist since the bvproducts of photoresist bombarded by the highenergy electron beam used to excite the Auger electrons is mostly carbon. As Table 1 illustrates, the plasma cleaning process was very effective in reducing this carbon contamination. Other authors have shown similar difficulty in thermocompression bonding gold lead frames to a gold circuit with more than 15 A of carbon present. The ionsputtered surfaces are listed for reference and verify a clean interior metallization. The small amount of carbon is probably residue from the ion sputtering. The other feature of this analysis was the.4es data off the bonding pads. This data illustrated that all the contamination was restricted to the bond pads. If a liftoff technique was used to passivate the chip, then photoresist applied to the bonding pads prior to passivation deposition would be baked on in such a manner that normal chemical removal would be difficult. Today this process has mostly been replaced by more efficient ones; however, it is not completely obsolete. Conclusions This study characterized a cleaning technique that is inexpensive and easy to employ as a prebond step. Specifically, the study concluded: 1) Plasma cleaning is an effective technique for removing the surface contaminations. detrimental to thermocompression bonding. Reprinted from the February 1979 issue of: wng 2) A parametric envelope was generated that shows the acceptable range of cleaning parameters for gold, aluminum. and silver hybrid metallization. 3) Since the range of parametes applicable to cleaning hybrids is wide, the process should be easy to control in production. 4) Plasma cleaning may be characterized in terms of only two parametersenergy input and gas flow rate. 5) Tantalum nitride resistors are not significantly influenced by plasma cleaning within the range of cleaning parameters evaluated. 6) Most organic contamination can be removed without affecting underlying inorganic materials. References 1. J. L. Jellin, Effect of Surface Contamination on the Thermocompression Bondability of Gold. Proceedings of the Electronic Component Conference. 1975. 2. LFE Data Sheet, Bulletm No. 8203 PBl, LFE Corporation, Process Control Dimion, Waltham, Massachusetts. February 1974. 3. D. S. Peck. Practical Application of Accelerated Testing. 13th Proceedings of Reliabiiity Physics. 1975. 4. Graphical StatisticsAn Enpneenng Approach, Weshnghouse Engmeer, MarchMay 1950. 5. P. A. Holloway, Quantitative Analysls of the Influence of Contaminants on Thermocompression Bonding of Gold %A731099 (March 1974). 6. F. K. Kane and K. L. Mittal, Plasma Cleaning of Metal Surfaces. Joumal of Vacuum Scrence and Technology, Vol;tl, N*3 MayJurwi974. _